How CardPreGrading works
Four photos in. A clear, defensible verdict out. Computer-vision centring measurement plus a guided defect review you do yourself.
Step 1 - Setup
You start a scan from your dashboard. We ask three things: the card name, the set, and which side you'll be photographing first. You can capture in your browser, upload existing photos, or get a one-tap link sent to your phone if you're working from a desktop.
Step 2 - Capture
We need four shots to give you a defensible verdict:
- Front, straight on. The primary shot used for the centering measurement.
- Back, straight on. Used to spot whitening and back surface issues.
- Front, angled. A second front shot at a slight angle to reveal surface scratches and holo damage that disappear under flat lighting.
- Four corners. A close-up that lets us zoom in to each corner during the defect review.
Before each photo is accepted we check it for blur, exposure, glare, and whether a card is actually in frame. If any check fails, we ask you to retake. Credits are not deducted until you have a valid set of four photos.
Step 3 - Process
Once your shots upload, we run a deterministic centering pipeline server-side. The card border is detected by contour analysis. The inner artwork frame is detected by a Sobel gradient sweep along each side. The L/R and T/B margins are computed and rounded to one decimal place.
Centering ratios map to PSA bands as follows:
- 50/50 → 55/45: PSA 10 centering threshold met.
- 55/45 → 60/40: PSA 9 ceiling.
- 60/40 → 65/35: PSA 8 ceiling.
- 65/35 → 70/30: PSA 7 ceiling.
- worse than 70/30: PSA 6 or below.
Step 4 - Report
You get the centering measurements, a guided defect review with zoom views, and a predicted grade range with a confidence score. The verdict - submit, borderline, or skip - translates the prediction into action.
How our AI works
CPG combines two layers. The first is a deterministic computer-vision pipeline that detects the card border and inner artwork frame, then computes centring ratios to 0.1mm. We chose a geometric algorithm over a learned model for centring because the problem has a closed-form answer - and because we want every measurement to be auditable. You can read the maths in our open methodology.
The second layer is the guided defect review with annotated zoom views. You tell us what you see; we apply PSA's published thresholds to translate your observations into a predicted grade range and verdict. As we collect more user-labelled data, we'll train ML models for surface and corner defect detection - but only where they outperform the guided checklist on real cards.
Multi-grader by design
Centring thresholds map to PSA's published scale by default - but the same measurement is portable to ACE, BGS, and CGC, all of which use comparable centring bands. CPG works whether you're submitting to PSA in the US, ACE in the UK, BGS for sports, or CGC for bulk.
What it can't do
We are not a guarantee. Human grading involves human judgement: a sub-grader on the day, their lighting, their threshold for a printer dot. CPG is the best informed second opinion you can get before you spend the $25+ submission fee - but it is not the actual grade.
Want to see what a finished report looks like? View the sample report →